mirror of
https://github.com/KhronosGroup/glslang
synced 2024-11-09 20:10:06 +00:00
257 lines
7.1 KiB
Plaintext
257 lines
7.1 KiB
Plaintext
#version 310 es
|
|
|
|
layout(local_size_x = 2) in;
|
|
layout(local_size_x = 16) in; // ERROR, changing
|
|
layout(local_size_z = 4096) in; // ERROR, too large
|
|
layout(local_size_x = 2) in;
|
|
layout(local_size_y = 0) in; // ERROR, 0 not allowed
|
|
const int total = gl_MaxComputeWorkGroupCount.y
|
|
+ gl_MaxComputeUniformComponents
|
|
+ gl_MaxComputeTextureImageUnits
|
|
+ gl_MaxComputeImageUniforms
|
|
+ gl_MaxComputeAtomicCounters
|
|
+ gl_MaxComputeAtomicCounterBuffers;
|
|
|
|
buffer ShaderStorageBlock
|
|
{
|
|
int value;
|
|
float values[];
|
|
};
|
|
|
|
buffer InvalidShaderStorageBlock
|
|
{
|
|
float values[]; // ERROR
|
|
int value;
|
|
} invalid;
|
|
|
|
void main()
|
|
{
|
|
barrier();
|
|
memoryBarrier();
|
|
memoryBarrierAtomicCounter();
|
|
memoryBarrierBuffer();
|
|
memoryBarrierShared();
|
|
memoryBarrierImage();
|
|
groupMemoryBarrier();
|
|
value = int(values[gl_LocalInvocationIndex]);
|
|
}
|
|
|
|
layout(location = 2) in vec3 v3; // ERROR
|
|
in float f; // ERROR
|
|
out float fo; // ERROR
|
|
|
|
shared vec4 s;
|
|
layout(location = 2) shared vec4 sl; // ERROR
|
|
shared float fs = 4.2; // ERROR
|
|
|
|
layout(local_size_x = 2, local_size_y = 3, local_size_z = 4) out; // ERROR
|
|
|
|
int arrX[gl_WorkGroupSize.x];
|
|
int arrY[gl_WorkGroupSize.y];
|
|
int arrZ[gl_WorkGroupSize.z];
|
|
|
|
readonly buffer roblock
|
|
{
|
|
int value;
|
|
float values[];
|
|
} ro;
|
|
|
|
void foo()
|
|
{
|
|
ro.values[2] = 4.7; // ERROR, readonly
|
|
ro.values.length();
|
|
++s;
|
|
}
|
|
|
|
buffer vec4 v; // ERROR
|
|
|
|
uniform usampler2D us2dbad; // ERROR, default precision
|
|
|
|
precision highp usampler2D;
|
|
precision highp iimage2DArray;
|
|
precision highp iimage2D;
|
|
|
|
uniform usampler2D us2d;
|
|
|
|
uniform iimage2DArray ii2dabad; // ERROR, not writeonly
|
|
uniform writeonly iimage2DArray ii2da;
|
|
|
|
layout(r32i) uniform iimage2D iimg2D;
|
|
layout(rgba32i) uniform readonly iimage2D iimg2Drgba;
|
|
layout(rgba32f) uniform readonly image2D img2Drgba; // ERROR, no default
|
|
layout(r32ui) uniform uimage2D uimg2D; // ERROR, no default
|
|
|
|
void qux()
|
|
{
|
|
int i = 4;
|
|
imageAtomicCompSwap(iimg2D, ivec2(i,i), i, i);// ERROR no longer in 310
|
|
imageAtomicAdd(uimg2D, ivec2(i,i), uint(i)); // ERROR no longer in 310
|
|
imageAtomicMin(iimg2Drgba, ivec2(i,i), i); // ERROR no longer in 310 // ERROR iimg2Drgba does not have r32i layout
|
|
imageAtomicMax(img2Drgba, ivec2(i,i), i); // ERROR no longer in 310 // ERROR img2Drgba is not integer image
|
|
ivec4 pos = imageLoad(iimg2D, ivec2(i,i));
|
|
imageStore(ii2da, ivec3(i,i,i), ivec4(0));
|
|
imageLoad(img2Drgba, ivec2(i,i));
|
|
imageLoad(ii2da, ivec3(i,i,i)); // ERROR, drops writeonly
|
|
}
|
|
|
|
volatile float vol; // ERROR, not an image
|
|
readonly int vol2; // ERROR, not an image
|
|
|
|
void passr(coherent readonly iimage2D image)
|
|
{
|
|
}
|
|
|
|
layout(r32i) coherent readonly uniform iimage2D qualim1;
|
|
layout(r32i) coherent restrict readonly uniform iimage2D qualim2;
|
|
|
|
void passrc()
|
|
{
|
|
passr(qualim1); // ERROR, changing formats
|
|
passr(qualim2); // ERROR, drops restrict, ERROR, changing formats
|
|
passr(iimg2D); // ERROR, changing formats
|
|
}
|
|
|
|
highp layout(rg8i) uniform readonly uimage2D i1bad; // ERROR, type mismatch
|
|
highp layout(rgba32i) uniform readonly image2D i2bad; // ERROR, type mismatch
|
|
highp layout(rgba32f) uniform readonly uimage2D i3bad; // ERROR, type mismatch
|
|
layout(r8_snorm) uniform readonly iimage2D i4bad; // ERROR, type mismatch
|
|
layout(rgba32ui) uniform readonly iimage2D i5bad; // ERROR, type mismatch
|
|
layout(r8ui) uniform readonly iimage2D i6bad; // ERROR, type mismatch
|
|
|
|
layout(binding = 0) uniform atomic_uint counter;
|
|
|
|
uint func(atomic_uint c)
|
|
{
|
|
return atomicCounterIncrement(c);
|
|
}
|
|
|
|
uint func2(out atomic_uint c) // ERROR, output
|
|
{
|
|
return counter; // ERROR, type mismatch
|
|
return atomicCounter(counter);
|
|
}
|
|
|
|
void mainAC()
|
|
{
|
|
atomic_uint non_uniform_counter; // ERROR
|
|
uint val = atomicCounter(counter);
|
|
atomicCounterDecrement(counter);
|
|
}
|
|
|
|
layout(binding = 1) uniform mediump atomic_uint counterBad; // ERROR, not highp
|
|
|
|
layout(binding = 2, offset = 4) uniform atomic_uint countArr[4];
|
|
uniform int i;
|
|
|
|
void opac()
|
|
{
|
|
int a[3];
|
|
a[counter]; // ERROR, non-integer
|
|
countArr[2];
|
|
countArr[i];
|
|
}
|
|
|
|
shared int atomi;
|
|
shared uint atomu;
|
|
|
|
void atoms()
|
|
{
|
|
int origi = atomicAdd(atomi, 3);
|
|
uint origu = atomicAnd(atomu, 7u);
|
|
origi = atomicExchange(atomi, 4);
|
|
origu = atomicCompSwap(atomu, 10u, 8u);
|
|
}
|
|
|
|
precision highp atomic_uint;
|
|
precision lowp atomic_uint; // ERROR
|
|
|
|
precise int pfoo; // ERROR, reserved
|
|
|
|
dmat2x4 dm; // ERROR
|
|
uniform samplerCubeArray sca; // ERROR
|
|
uniform iimage2DRect i2dr; // ERROR
|
|
highp uniform image2DMS i2dms; // ERROR
|
|
uniform uimage2DMSArray u2dmsa; // ERROR
|
|
|
|
highp layout(r32f) coherent volatile restrict readonly writeonly uniform image2D okay1;
|
|
layout(r32i) coherent volatile restrict readonly uniform iimage2D okay2;
|
|
highp layout(r32ui) coherent volatile restrict writeonly uniform uimage2D okay3;
|
|
highp layout(r32f) coherent volatile restrict uniform image2D okay4;
|
|
|
|
highp layout(rgba32f) coherent volatile restrict uniform image2D badQ1; // ERROR, bad qualifiers
|
|
layout(rgba8i) coherent volatile restrict uniform iimage2D badQ2; // ERROR, bad qualifiers
|
|
highp layout(rgba16ui) coherent volatile restrict uniform uimage2D badQ3; // ERROR, bad qualifiers
|
|
|
|
writeonly buffer woblock
|
|
{
|
|
int value;
|
|
float values[];
|
|
} wo;
|
|
|
|
void foowo()
|
|
{
|
|
float g;
|
|
g = wo.values[2]; // ERROR, writeonly
|
|
float f = wo.values[2]; // ERROR, writeonly
|
|
++wo.values[2]; // ERROR, writeonly
|
|
wo.values[2]--; // ERROR, writeonly
|
|
f + wo.values[2]; // ERROR, writeonly
|
|
wo.values[2] - f; // ERROR, writeonly
|
|
bool b;
|
|
b ? f : wo.values[2]; // ERROR, writeonly
|
|
b ? wo.values[2] : f; // ERROR, writeonly
|
|
if (f == wo.values[2]) // ERROR, writeonly
|
|
++f;
|
|
if (f >= wo.values[2]) // ERROR, writeonly
|
|
++f;
|
|
f = vec3(wo.values[2]).x; // ERROR, writeonly
|
|
~wo.value; // ERROR, writeonly
|
|
wo.values[2] = 3.4;
|
|
}
|
|
|
|
buffer multioblock
|
|
{
|
|
readonly int value;
|
|
writeonly float values[];
|
|
} multio;
|
|
|
|
void foomultio()
|
|
{
|
|
float g;
|
|
g = wo.values[2]; // ERROR, writeonly
|
|
~wo.value;
|
|
wo.values[2] = 3.4;
|
|
wo.value = 2; // ERROR, readonly
|
|
}
|
|
|
|
in inb { // ERROR
|
|
int a;
|
|
} inbi;
|
|
|
|
out outb { // ERROR
|
|
int a;
|
|
} outbi;
|
|
|
|
float t__; // ERROR, no __ until revision 310
|
|
|
|
// ERROR, no __ until revision 310
|
|
#define __D
|
|
|
|
shared vec4 arr[2][3][4];
|
|
|
|
void devi()
|
|
{
|
|
gl_DeviceIndex; // ERROR, no extension
|
|
gl_ViewIndex; // ERROR, never this stage
|
|
}
|
|
|
|
#ifdef GL_EXT_device_group
|
|
#extension GL_EXT_device_group : enable
|
|
#endif
|
|
|
|
void devie()
|
|
{
|
|
gl_DeviceIndex;
|
|
gl_ViewIndex; // ERROR, never this stage
|
|
}
|